Handwritten Chinese Character Recognition by Guided Stroke Extraction and Hidden Markov Models

摘要

This paper presents an approach to the recognition of handwritten Chinese characters by using guided stroke extraction, vector quantization and hidden Markov modeling techniques. It targets on the final recognition stage which is supposed to , distinguish characters in a relative small character set from pre-classifiers. Different aspect of modeling Chinese character by HMMs will be discussed. In our preliminary tests, LBG algorithm is used in vector quantization, Forward-Backward procedure and Baum-Welch re-estimation algorithm are used in training phase, and the Viterbi algorithm is used for recognition. Encouraging results of over 95% of recognition rate is obtained and there are still potential for further improvement.